Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76347.98179285588,
-      "required_pages_per_second_upper_estimate": 0.0052819305321542736
+      "required_token_positions_per_second": 77359.37672243846,
+      "required_pages_per_second_upper_estimate": 0.005351901180142354
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9543.497724106985,
-      "required_pages_per_second_upper_estimate": 0.003995663206188992
+      "required_token_positions_per_second": 9669.922090304808,
+      "required_pages_per_second_upper_estimate": 0.0040485944482750455
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19086.99544821397,
-      "required_pages_per_second_upper_estimate": 0.003975317793943646
+      "required_token_positions_per_second": 19339.844180609616,
+      "required_pages_per_second_upper_estimate": 0.004027979516832178
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28630.493172320952,
-      "required_pages_per_second_upper_estimate": 0.0017163543494488189
+      "required_token_positions_per_second": 29009.766270914424,
+      "required_pages_per_second_upper_estimate": 0.0017390911925930078
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28630.493172320952,
-      "required_pages_per_second_upper_estimate": 0.0022258949119174174
+      "required_token_positions_per_second": 29009.766270914424,
+      "required_pages_per_second_upper_estimate": 0.0022553817270870046
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28630.493172320952,
-      "required_pages_per_second_upper_estimate": 0.36724122538603854
+      "required_token_positions_per_second": 29009.766270914424,
+      "required_pages_per_second_upper_estimate": 0.3721061334630703
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2774258024687697,
-    "producer_p95_seconds": 0.35956035800336394,
-    "gpu_update_seconds": 0.9656836797603334,
-    "producer_batches_per_second": 3.604567387392064,
-    "gpu_batches_per_second": 1.0355357773553586,
-    "producer_headroom": 3.4808718986008595,
+    "producer_batch_seconds": 0.12412315800793294,
+    "producer_p95_seconds": 0.34494191599515034,
+    "gpu_update_seconds": 0.9530583508258131,
+    "producer_batches_per_second": 8.056514320527425,
+    "gpu_batches_per_second": 1.0492536990348098,
+    "producer_headroom": 7.6783282517264135,
     "sustainable": true,
-    "buffer_seconds": 60.83807182490101,
+    "buffer_seconds": 60.04267610202623,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 305890.22085250507,
+    "generation_bytes_per_second": 306428.64746043325,
     "upload_bytes_per_second": 4611187.090552738,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.63692063504566,
+    "estimated_full_shards_per_hour": 65.7524544511771,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -408,3 +408,4 @@
 {"step": 800800, "loss": 0.5392630696296692, "lr": 3e-05, "grad_norm": 0.2653088635396031, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117213898762, "tokens_per_second": 141825.97205698508, "data_wait_s": 1.095700281439349e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 800900, "loss": 0.548346757888794, "lr": 3e-05, "grad_norm": 0.2749633752369359, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117228365568, "tokens_per_second": 149780.98371397235, "data_wait_s": 6.10799907008186e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 801000, "loss": 0.5692843794822693, "lr": 3e-05, "grad_norm": 0.2703526957225969, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117242932024, "tokens_per_second": 151489.24277070843, "data_wait_s": 9.110997780226171e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 801100, "loss": 0.5213542580604553, "lr": 3e-05, "grad_norm": 0.2853517031856212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117257534817, "tokens_per_second": 151063.8605249472, "data_wait_s": 1.109600270865485e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9b92025dd58f58396e0695da71401db0688ab9ce4b18bd68eec4c558d83ed7a0
-size 874809
+oid sha256:d2460f4f940be382f63a8a2c54b5bb560675175016bdd57298bc4eca481e62ad
+size 876927